So many groups are about to find that their software program practices are worse than they thought. Undocumented scripts, magical native setup, flaky checks, tribal-knowledge structure, imprecise tickets, inconsistent naming, and “each senior engineer does it a bit in another way.” People simply discovered to soak up it. Brokers expose this silliness instantly. An underspecified surroundings doesn’t create creativity; it creates rubbish. If you happen to drop an agent right into a messy codebase and it flails, that’s not essentially an indictment of the agent. Usually it’s a really environment friendly audit of your engineering self-discipline. The repo is lastly telling the reality about itself.
Which is why I’d now say that my suggestion that AI coding requires builders to turn into higher managers was true, if incomplete. Sure, builders must turn into higher managers of machines. However extra importantly, they should turn into higher engineers within the old style sense: higher at specs, boundaries, “golden paths,” and so on. The agent period rewards self-discipline excess of cleverness, and that’s most likely overdue.
So no, the large story of coding brokers isn’t that they’ll write code. Plain chatbots may already pretend that half. The large story is that they’re altering what competent software program engineering seems to be like. Brokers reward precisely the issues builders have lengthy claimed to worth however typically averted in apply: explicitness, consistency, testability, and proof. Within the age of brokers, boring software program engineering doesn’t simply scale higher, it does most every part—collaboration, debugging, and so on.—higher.
